To enhance the accuracy of corrosion and breakpoint quantification in grounding grids, an inversion method based on transient electromagnetic apparent resistivity was proposed by combining the IVY optimization algorithm with a convolutional neural networks-bidirectional long short-term memory(CNN-BiLSTM) deep learning model. Through this approach, high-precision inversion of transient electromagnetic signals from grounding grids was achieved. Traditional transient electromagnetic inversion methods, such as the “smoke ring” theory, enable rapid imaging but suffer from limited inversion accuracy, while conventional CNN-BiLSTM hybrid models can capture spatiotemporal features but lack effective hyperparameter optimization. The IVY algorithm was introduced to adaptively optimize the hyperparameters of the CNN-BiLSTM model, thereby significantly improving its performance. A one-dimensional forward simulation was conducted to construct a training dataset containing various geoelectrical parameters, and the IVY-CNN-BiLSTM hybrid model was employed to extract the deep mapping relationship between transient electromagnetic signals and apparent resistivity, realizing end-to-end high-precision inversion. The inversion results demonstrated that the proposed method achieved a coefficient of determination exceeding 0.95 with the true values and outperformed traditional CNN and CNN-BiLSTM algorithms. The proposed approach provides a reliable basis for nondestructive quantitative evaluation and visual diagnosis of grounding grid conditions. Experimental validation confirmed that the proposed method exhibited much higher accuracy and reliability in grounding grid corrosion diagnosis than conventional inversion methods, offering a novel technical route for substation grounding grid condition monitoring.